45–52,56,60–69], which include variable precision rough set models [56], rough set models based on tolerance relation
[12,39], fuzzy rough set models [5,7,11,21,41], fuzzy probabilistic rough set models [10], etc.
It should be noticed that in most of the extensions of the rough set models, the used binary relations induce a covering
instead of a partition on the universe of discourse, and then, it is natural to extend the classical rough set model to the cov-
ering environment. In other words, the covering rough set models [1,2,7–9,22,24,37,38,42,45,58–69] are important issues to
be addressed. In [37], Samanta presented sixteen covering rough set models and studied their implication lattices, six of
which were also been systematically studied by Zhu in [61–69].Thus, we take the six covering rough set models which were
deeply explored by the two scholars as examples to introduce the basic concepts about the generalized rough set in covering.
Zakowski extended Pawlak’s rough set theory from partition to covering and proposed the first covering rough set model in
1983 [55]. Following Zakowski’s work, Pomykala proposed the second covering rough set model in 1987 [31]. His main
method was the interior operator which is adopted by the topology theory. The third covering rough set model was first de-
fined in [42]. The definition of the third type of upper approximation operation is believed to be more reasonable than those
of the first and second types, but no properties of this new class of covering generalized rough sets have been discussed. By
combining the definitions of three types of covering rough set, Zhu and Wang proposed the fourth covering rough set model
in [61]. They presented the basic properties of the fourth covering rough set, and studied the independency between the low-
er and the upper approximations. From the topological point of view, Zhu presented the fifth covering rough set model in
[62], and explored the detailed properties of lower and upper approximations for this new type of rough sets. It’s worth not-
ing that the lower approximations of these five models are same. Then, in Ref. [63], the sixth covering rough set model was
defined by Zhu. This model includes not only the covering upper approximation but also the covering lower approximation.
Readers can obtain the characteristics of the covering rough set models through the above six models.
On the other hand, Lin presented a rough set model in neighborhood system from the viewpoint of the concepts of inte-
rior and closure in topology theory [1,14–23,26]. The generalized rough set in neighborhood system also becomes a hotspot
in current researches. In [6], Das established a necessary and sufficient condition for a group of fuzzy sets in a group to be the
family of neighborhoods of the identity element of a fuzzy topological group. Lin explored neighborhood systems and
approximation in relational database and knowledge bases in [17,26]. In those papers, approximate retrieval in database
was also initiated. Information retrieval was proposed in [18]. Approximation retrieval and information retrieval were sum-
marized in [15]. After Lin’s work, Yao did some researches about neighborhood system and approximate retrieval model
based on notions of neighborhood systems in [53]. In fact, most of the rough set models can be regarded as established
on the basis of different neighborhoods. For example, equivalence class is a neighborhood, the covering in covering approx-
imation space is also a neighborhood, and fuzzy equivalence class is a neighborhood, as well. Therefore, the relationships
between current rough set models and rough set model in neighborhood system are worthy to be discussed.
From the introduction above, we can find that the generalized rough sets in coverings and neighborhood system have
attracted interests of many researchers and practitioners in various fields of science and technology. Pure reflexive neighbor-
hood system is a special type of neighborhood system. All the neighborhoods are reflexive and have no empty set. This is
consistent with the requirement of covering. Naturally, we will ask the question ‘‘what are connections and differences be-
tween the two generalized rough set models?’’ This is an important and interesting question. This paper just aims to answer
it. The goal of this article is to find out the commonalities among these different rough set models and to grasp the devel-
opment direction of rough set theory by comparative analysis of the two generalized rough sets. Through the analysis of their
similarities and differences, a hand will be given when facing model selection problems. This paper can be seen as the further
exploration for these two generalized rough sets.
In this paper, we mainly focus on the generalized rough set in six coverings [24,61–69] and the generalized rough set in
pure reflexive neighborhood system. That is to say that total seven rough set models are systematically studied in this paper.
The major contributions of this paper are to find out the inclusion relations or equivalence relations among seven upper/low-
er approximations, and to obtain the relationships among seven accuracy measures. The results can be summarized as fol-
lows: among the six covering rough set models, the accuracy measures of the first, second, third and fifth are comparable.
Meanwhile, the accuracy measures of the first, second, fourth and fifth model are also comparable. However, there’s no rela-
tionship between the accuracy measures of the third and the fourth covering rough set models. The accuracy measure of the
sixth covering rough set model is lower than that of the second covering rough set model. However, it cannot be compared
with the accuracy measures of the other four covering rough set models. Then, between the generalized rough sets in pure
reflexive neighborhood system and covering, the accuracy measure of pure reflexive neighborhood system rough set model
and that of the second and sixth covering rough set model are comparable. However, it cannot be compared with the accu-
racy measures of the first, third, fourth and fifth covering rough set model. Thus, different rough set models may be chosen
according to different accuracy requirements.
The remainder of this paper is organized as follows. In Section 2, we briefly introduce the fundamental concepts of the
classical rough set. In addition, the basic notions of the generalized rough set in six coverings and that in pure reflexive
neighborhood system are been introduced. The major contributions of this paper are covered in Section 3 and 4. In Section 3,
the relationships among generalized rough sets in six coverings and pure reflexive neighborhood system are deeply ex-
plored. It includes two aspects: relationships among six covering rough set models are studied in Section 3.1, relationships
between generalized rough sets in pure reflexive neighborhood system and each covering are explored in Section 3.2. In Sec-
tion 4, the relationships among the accuracy measures of all seven rough set models are acquired. Results are summarized in
Section 5.
L. Wang et al. / Information Sciences 207 (2012) 66–78
67